Minutes_2018_08_30

Numba Meeting: 2018-08-30

Attendees: Todd, Ehsan, Siu, Stuart

What to do about #3195?
- "Problems with support_literals and the Const type"
- Problems:
  - Lowering backend mishandle Const type
    - likely caused other arguments that are not expecting a Const
  - Blocking:
    - np.dtype(str) and other str usage
- Suggested Solution:
  1. autoconversion of Const into bare Type
    - Siu: try this by next meeting
  2. support_literals for particular argument by name or position
  3. QualType or metadata; remove Const type
Status and plans for threading backend work
- Status:
  - current threading pool not threadsafe
  - replaced with parallel_for loop
  - 2 backends: TBB & OMP
  - merged+refactored gufunc and ufunc
  - Backend picking; i.e. fork-safe backend, ...
- Problems:
  - fork+exec issues. {fork, thread}-safety different on every platform; incl python vers (esp. 2.7) and OS vers.
  - linux is most problematic, surprisingly
- Suggestions:
  - blacklist parallel support on certain platforms
- Moving forward:
  - Is there a backend that is better than the current one?
    - Ans: There's one
  - TBB won't be default or strict dependency yet.
- Bumping release later by a week to give more time to get threading backend stabilized
Parallel Accelerator to graduate from being experimental?
- Change doc to indicate PA is no longer experimental.
- But, it is still not on by default, yet.

#3279 - test_fuse_argmin unexpected IR in lowering
- Sporadic failure. Likely ordering of test is causing it.
- Blocker.
#3277 - autojit crashed unexpectedly
- Unsupported. Old versions.
#3271 - "Numba requires at least version 0.25.0 of llvmlite." (current is 0.24)
- Pending user response.
#3267 - Automatic detection of changes to globals at call time to trigger recompilation
- feature request.
- nothing to do yet
#3265 - Unexpected results I am getting. BUG in the CUDA Simulator.
- "bug" is Numba behavior regarding freezing globals in CUDA code vs. direct access to globals in normal Python
#3263 - Document known memory leak associated with exceptions
#3262 - Memory leak when exception occurs in np.linalg.svd bug duplicate

3281 - Fix ds_permute types and add tests
3279 -compiler threadsafety lockdown
3274 -[WIP] Update installation instructions
- WIP because instructions assume ARMv7 packages in Numba channel
- Otherwise ready for review
3273 -Refactor. Removing thread-local-storage based context nesting.
3272 - Add MemInfo_data to runtime._nrt_python.c_helpers
3259 - Add DeviceNDArray.view method to reinterpret data as a different type.
- New tests need to skip CUDA simulator because they assume device array methods

3234 - Fix types and layout for np.where.
- Siu will review
3228 - Reduce redundant module linking
- Stuart will review
3212 - Support for np.vander
- Stuart will review
3211 - Handle unpacking in building tuple (BUILD_TUPLE_UNPACK opcode)
- Reviewed
- Need to verify Mac works
- Siu will merge
3209 - Support for np.tri, np.tril and np.triu
- blocked on #3234
- author on vacation
3202 - [WIP] TBB + backend refactor... DO NOT MERGE!
- Currently chasing different bugs on different platforms
3186 Support Records in CUDA Const Memory
- Stan needs to get back to user with PTX on failing systems
3172 Use float64 add Atomics, Where Available
- needs re-review by Siu
3166 [WIP] Objmode with-block
- Status?
3162 Support constant dtype string in nopython mode in functions like numpy.empty.
- Need to resolve #3195
3160 First attempt at parallel diagnostics
- Stuart will implement Todd's suggestion
3142 Issue3139
- Blocked on 3127
3134 [WIP] Cfunc x86 abi
- Needs re-review
3124 Fix 3119, raise for 0d arrays in reductions
- Stuart needs to implement feedback
3093 [WIP] Singledispatch overload support for cuda array interface.
- Needs review
3046 Pairwise sum implementation.
#2999 Support LowLevelCallable
#2983 [WIP] invert mapping b/w binop operators and the operator module
#2950 Fix dispatcher to only consider contiguous-ness.
#2942 Fix linkage nature (declspec(dllexport)) of some test functions
#2894: [WIP] Implement jitclass default constructor arguments.
#2817: [WIP] Emit LLVM optimization remarks

===========================